Saved in:
| Main Authors: | , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2026
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2603.24477 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866908913738186752 |
|---|---|
| author | Research, Cursor : Chan, Aaron Shalaby, Ahmed Wettig, Alexander Sanger, Aman Zhai, Andrew Ajay, Anurag Nair, Ashvin Snell, Charlie Lu, Chen Shen, Chen Jia, Emily Cassano, Federico Liu, Hanpeng Chen, Haoyu Wildermuth, Henry Jackson, Jacob Li, Janet Katz, Jediah Yao, Jiajun Hejna, Joey Warner, Josh Vering, Julius Frans, Kevin Danilek, Lee Wright, Less Cen, Lujing Melas-Kyriazi, Luke Truell, Michael de Jong, Michiel Jain, Naman Schmidt, Nate Wang, Nathan Muennighoff, Niklas Rybkin, Oleg Loh, Paul Kravtsov, Phillip Yadav, Rishabh Shah, Sahil Kottler, Sam Rush, Alexander M Zhang, Shengtong Jain, Shomil Sankar, Sriram Heule, Stefan Sul, Stuart H. Asif, Sualeh Rong, Victor Zhu, Wanqi Lin, William Wu, Yuchen Volkov, Yuri Zemlyanskiy, Yury Holbrook, Zack Zhang, Zhiyuan |
| author_facet | Research, Cursor : Chan, Aaron Shalaby, Ahmed Wettig, Alexander Sanger, Aman Zhai, Andrew Ajay, Anurag Nair, Ashvin Snell, Charlie Lu, Chen Shen, Chen Jia, Emily Cassano, Federico Liu, Hanpeng Chen, Haoyu Wildermuth, Henry Jackson, Jacob Li, Janet Katz, Jediah Yao, Jiajun Hejna, Joey Warner, Josh Vering, Julius Frans, Kevin Danilek, Lee Wright, Less Cen, Lujing Melas-Kyriazi, Luke Truell, Michael de Jong, Michiel Jain, Naman Schmidt, Nate Wang, Nathan Muennighoff, Niklas Rybkin, Oleg Loh, Paul Kravtsov, Phillip Yadav, Rishabh Shah, Sahil Kottler, Sam Rush, Alexander M Zhang, Shengtong Jain, Shomil Sankar, Sriram Heule, Stefan Sul, Stuart H. Asif, Sualeh Rong, Victor Zhu, Wanqi Lin, William Wu, Yuchen Volkov, Yuri Zemlyanskiy, Yury Holbrook, Zack Zhang, Zhiyuan |
| contents | Composer 2 is a specialized model designed for agentic software engineering. The model demonstrates strong long-term planning and coding intelligence while maintaining the ability to efficiently solve problems for interactive use. The model is trained in two phases: first, continued pretraining to improve the model's knowledge and latent coding ability, followed by large-scale reinforcement learning to improve end-to-end coding performance through stronger reasoning, accurate multi-step execution, and coherence on long-horizon realistic coding problems. We develop infrastructure to support training in the same Cursor harness that is used by the deployed model, with equivalent tools and structure, and use environments that match real problems closely. To measure the ability of the model on increasingly difficult tasks, we introduce a benchmark derived from real software engineering problems in large codebases including our own. Composer 2 is a frontier-level coding model and demonstrates a process for training strong domain-specialized models. On our CursorBench evaluations the model achieves a major improvement in accuracy compared to previous Composer models (61.3). On public benchmarks the model scores 61.7 on Terminal-Bench and 73.7 on SWE-bench Multilingual in our harness, comparable to state-of-the-art systems. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2603_24477 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Composer 2 Technical Report Research, Cursor : Chan, Aaron Shalaby, Ahmed Wettig, Alexander Sanger, Aman Zhai, Andrew Ajay, Anurag Nair, Ashvin Snell, Charlie Lu, Chen Shen, Chen Jia, Emily Cassano, Federico Liu, Hanpeng Chen, Haoyu Wildermuth, Henry Jackson, Jacob Li, Janet Katz, Jediah Yao, Jiajun Hejna, Joey Warner, Josh Vering, Julius Frans, Kevin Danilek, Lee Wright, Less Cen, Lujing Melas-Kyriazi, Luke Truell, Michael de Jong, Michiel Jain, Naman Schmidt, Nate Wang, Nathan Muennighoff, Niklas Rybkin, Oleg Loh, Paul Kravtsov, Phillip Yadav, Rishabh Shah, Sahil Kottler, Sam Rush, Alexander M Zhang, Shengtong Jain, Shomil Sankar, Sriram Heule, Stefan Sul, Stuart H. Asif, Sualeh Rong, Victor Zhu, Wanqi Lin, William Wu, Yuchen Volkov, Yuri Zemlyanskiy, Yury Holbrook, Zack Zhang, Zhiyuan Software Engineering Machine Learning Composer 2 is a specialized model designed for agentic software engineering. The model demonstrates strong long-term planning and coding intelligence while maintaining the ability to efficiently solve problems for interactive use. The model is trained in two phases: first, continued pretraining to improve the model's knowledge and latent coding ability, followed by large-scale reinforcement learning to improve end-to-end coding performance through stronger reasoning, accurate multi-step execution, and coherence on long-horizon realistic coding problems. We develop infrastructure to support training in the same Cursor harness that is used by the deployed model, with equivalent tools and structure, and use environments that match real problems closely. To measure the ability of the model on increasingly difficult tasks, we introduce a benchmark derived from real software engineering problems in large codebases including our own. Composer 2 is a frontier-level coding model and demonstrates a process for training strong domain-specialized models. On our CursorBench evaluations the model achieves a major improvement in accuracy compared to previous Composer models (61.3). On public benchmarks the model scores 61.7 on Terminal-Bench and 73.7 on SWE-bench Multilingual in our harness, comparable to state-of-the-art systems. |
| title | Composer 2 Technical Report |
| topic | Software Engineering Machine Learning |
| url | https://arxiv.org/abs/2603.24477 |